v0.27.8: Zero-Assumption Interactive Interview, Native Backlog Support & Host Sandbox Security

Zero-Assumption Interactive Interview, Native Backlog Support & Host Sandbox Security

• Zero-Assumption & Missing-Scenario Interview Invariant (harness-rules.md, AGENTS.md, pre_invocation_reminder.py):
  • Strictly prohibited AI agents from guessing, assuming, or inventing business logic, UI error texts, or missing scenarios from their own heads.
  • Mandated that agents audit for unaddressed network states (offline, timeout), state invariants (empty country/ISO), caching TTL, and error handling after graph exploration.
  • Required proactive developer interviews via ask_question with structured choices before authoring implementation_plan.md, ensuring plans are agreed upon and correct from the first attempt.
• Attached Media First-Turn Inspection Invariant (harness-rules.md, AGENTS.md, new-feature.md):
  • Enforced that whenever the developer provides attached screenshots, images, or screen recordings, the Lead Agent MUST view and inspect the media via view_file in the very first turn before searching code or authoring plans.
• Native Zoho Sprints Backlog Resolution (agents/mcp/zoho_sprints/_client.py):
  • Added native get_backlog_id() via /?action=getbacklog API query and unified backlog caching.
  • Expanded resolve_item() to automatically search both active/future sprints and the project backlog, enabling sub-second resolution of backlog items (e.g. I769, I770) without ID mismatches.
• Host OS Traversal & Anti-Scraping Sandbox Guard (pre_tool_safety.py):
  • Implemented fail-closed interception denying shell commands that attempt to scan host user home directories (C:\Users\..., /home/..., ~, %USERPROFILE%).
  • Added strict Fail-Fast Tracker Policy: maximum 1 attempt for issue lookup; hard deny on reverse-engineering, Google searches for internal APIs, or scratch scrapers.

Full Changelog: v0.27.7...v0.27.8